Lip segmentation based on Lambertian shadings and morphological operators for hyper-spectral images. (March 2017)

• Record Type: Journal Article
• Authors: Danielis, Alessandro; Giorgi, Daniela; Larsson, Marcus; Strömberg, Tomas; Colantonio, Sara; Salvetti, Ovidio
• Is Part Of: Pattern recognition. Volume 63(2017:Mar.)
• Page Start: 355
• Page End: 370
• Publication Date: 2017-03
• Subjects: Lip spatial pattern -- Segmentation -- Blood concentration map -- Hyper-spectral -- Lambertian shading -- Morphological -- Fourier descriptors; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.10.007
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Lip segmentation is a non-trivial task because the colour difference between the lip and the skin regions maybe not so noticeable sometimes. We propose an automatic lip segmentation technique for hyper-spectral images from an imaging prototype with medical applications. Contrarily to many other existing lip segmentation methods, we do not use colour space transformations to localise the lip area. As input image, we use for the first time a parametric blood concentration map computed by using narrow spectral bands. Our method mainly consists of three phases: (i) for each subject generate a subset of face images enhanced by different simulated Lambertian illuminations, then (ii) perform lip segmentation on each enhanced image by using constrained morphological operations, and finally (iii) extract features from Fourier-based modeled lip boundaries for selecting the lip candidate. Experiments for testing our approach are performed under controlled conditions on volunteers and on a public hyper-spectral dataset. Results show the effectiveness of the algorithm against low spectral range, moustache, and noise. Abstract : Highlights: We propose an automatic lip segmentation method for hyper-spectral images using a blood concentration map as input data. We generate a subset of Lambertian shaded maps per input image as a pre-processing step to enhance the lip spatial pattern. We perform morphological reconstructions under texture constraints for locating the lip region in each subset. We use Fourier descriptors to model reconstructed and shaded lip boundaries and to extract the segmented candidate. The experiments have been conducted on hyper-spectral images from a private and a public datasets.